1 /* $OpenBSD: if_pfsync.c,v 1.73 2006/11/16 13:13:38 henning Exp $ */ 2 3 /* 4 * Copyright (c) 2002 Michael Shalayeff 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 20 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 24 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 25 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 26 * THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include "opt_inet.h" 30 #include "opt_inet6.h" 31 #include "use_carp.h" 32 33 #include <sys/param.h> 34 #include <sys/endian.h> 35 #include <sys/proc.h> 36 #include <sys/priv.h> 37 #include <sys/systm.h> 38 #include <sys/time.h> 39 #include <sys/mbuf.h> 40 #include <sys/socket.h> 41 #include <sys/kernel.h> 42 #include <sys/malloc.h> 43 #include <sys/module.h> 44 #include <sys/sockio.h> 45 #include <sys/thread2.h> 46 #include <vm/vm_zone.h> 47 48 #include <machine/inttypes.h> 49 50 #include <net/if.h> 51 #include <net/if_types.h> 52 #include <net/route.h> 53 #include <net/bpf.h> 54 #include <netinet/in.h> 55 #include <netinet/if_ether.h> 56 #include <netinet/ip_carp.h> 57 #include <netinet/tcp.h> 58 #include <netinet/tcp_seq.h> 59 60 #ifdef INET 61 #include <netinet/in_systm.h> 62 #include <netinet/in_var.h> 63 #include <netinet/ip.h> 64 #include <netinet/ip_var.h> 65 #endif 66 67 #ifdef INET6 68 #include <netinet6/nd6.h> 69 #endif /* INET6 */ 70 71 #include <net/pf/pfvar.h> 72 #include <net/pf/if_pfsync.h> 73 74 #define PFSYNCNAME "pfsync" 75 76 #define PFSYNC_MINMTU \ 77 (sizeof(struct pfsync_header) + sizeof(struct pf_state)) 78 79 #ifdef PFSYNCDEBUG 80 #define DPRINTF(x) do { if (pfsyncdebug) kprintf x ; } while (0) 81 int pfsyncdebug; 82 #else 83 #define DPRINTF(x) 84 #endif 85 86 struct pfsync_softc *pfsyncif = NULL; 87 struct pfsyncstats pfsyncstats; 88 89 void pfsyncattach(int); 90 static int pfsync_clone_destroy(struct ifnet *); 91 static int pfsync_clone_create(struct if_clone *, int, caddr_t); 92 void pfsync_setmtu(struct pfsync_softc *, int); 93 int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 94 struct pf_state_peer *); 95 int pfsync_insert_net_state(struct pfsync_state *, u_int8_t); 96 int pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *, 97 struct rtentry *); 98 int pfsyncioctl(struct ifnet *, u_long, caddr_t, struct ucred *); 99 void pfsyncstart(struct ifnet *); 100 101 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **); 102 int pfsync_request_update(struct pfsync_state_upd *, struct in_addr *); 103 int pfsync_sendout(struct pfsync_softc *); 104 int pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *); 105 void pfsync_timeout(void *); 106 void pfsync_send_bus(struct pfsync_softc *, u_int8_t); 107 void pfsync_bulk_update(void *); 108 void pfsync_bulkfail(void *); 109 110 static MALLOC_DEFINE(M_PFSYNC, PFSYNCNAME, "Packet Filter State Sync. Interface"); 111 static LIST_HEAD(pfsync_list, pfsync_softc) pfsync_list; 112 113 int pfsync_sync_ok; 114 115 struct if_clone pfsync_cloner = 116 IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy, 1 ,1); 117 118 void 119 pfsyncattach(int npfsync) 120 { 121 if_clone_attach(&pfsync_cloner); 122 } 123 static int 124 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param __unused) 125 { 126 struct pfsync_softc *sc; 127 struct ifnet *ifp; 128 129 lwkt_gettoken(&pf_token); 130 131 MALLOC(sc, struct pfsync_softc *, sizeof(*sc), M_PFSYNC, 132 M_WAITOK|M_ZERO); 133 134 pfsync_sync_ok = 1; 135 sc->sc_mbuf = NULL; 136 sc->sc_mbuf_net = NULL; 137 sc->sc_mbuf_tdb = NULL; 138 sc->sc_statep.s = NULL; 139 sc->sc_statep_net.s = NULL; 140 sc->sc_statep_tdb.t = NULL; 141 sc->sc_maxupdates = 128; 142 sc->sc_sync_peer.s_addr =htonl(INADDR_PFSYNC_GROUP); 143 sc->sc_sendaddr.s_addr = htonl(INADDR_PFSYNC_GROUP); 144 sc->sc_ureq_received = 0; 145 sc->sc_ureq_sent = 0; 146 sc->sc_bulk_send_next = NULL; 147 sc->sc_bulk_terminator = NULL; 148 149 lwkt_reltoken(&pf_token); 150 ifp = &sc->sc_if; 151 ksnprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit); 152 if_initname(ifp, ifc->ifc_name, unit); 153 ifp->if_ioctl = pfsyncioctl; 154 ifp->if_output = pfsyncoutput; 155 ifp->if_start = pfsyncstart; 156 ifp->if_type = IFT_PFSYNC; 157 ifp->if_snd.ifq_maxlen = ifqmaxlen; 158 ifp->if_hdrlen = PFSYNC_HDRLEN; 159 ifp->if_baudrate = IF_Mbps(100); 160 ifp->if_softc = sc; 161 pfsync_setmtu(sc, MCLBYTES); 162 callout_init(&sc->sc_tmo); 163 callout_init(&sc->sc_tdb_tmo); 164 callout_init(&sc->sc_bulk_tmo); 165 callout_init(&sc->sc_bulkfail_tmo); 166 if_attach(&sc->sc_if, NULL); 167 168 LIST_INSERT_HEAD(&pfsync_list, sc, sc_next); 169 bpfattach(&sc->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN); 170 171 #if NCARP > 0 172 if_addgroup(ifp, "carp"); 173 #endif 174 lwkt_gettoken(&pf_token); 175 176 lwkt_reltoken(&pf_token); 177 return (0); 178 } 179 180 static int 181 pfsync_clone_destroy(struct ifnet *ifp) 182 { 183 lwkt_gettoken(&pf_token); 184 lwkt_reltoken(&pf_token); 185 #if NBPFILTER > 0 186 bpfdetach(ifp); 187 #endif 188 if_detach(ifp); 189 lwkt_gettoken(&pf_token); 190 kfree(pfsyncif, M_DEVBUF); 191 pfsyncif = NULL; 192 193 return 0; 194 } 195 196 /* 197 * Start output on the pfsync interface. 198 */ 199 void 200 pfsyncstart(struct ifnet *ifp) 201 { 202 crit_enter(); 203 IF_DROP(&ifp->if_snd); 204 IF_DRAIN(&ifp->if_snd); 205 crit_exit(); 206 } 207 208 int 209 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 210 struct pf_state_peer *d) 211 { 212 if (s->scrub.scrub_flag && d->scrub == NULL) { 213 d->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT); 214 if (d->scrub == NULL) 215 return (ENOMEM); 216 bzero(d->scrub, sizeof(*d->scrub)); 217 } 218 219 return (0); 220 } 221 222 int 223 pfsync_insert_net_state(struct pfsync_state *sp, u_int8_t chksum_flag) 224 { 225 struct pf_state *st = NULL; 226 struct pf_state_key *sk = NULL; 227 struct pf_rule *r = NULL; 228 struct pfi_kif *kif; 229 230 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) { 231 kprintf("pfsync_insert_net_state: invalid creator id:" 232 " %08x\n", ntohl(sp->creatorid)); 233 return (EINVAL); 234 } 235 236 kif = pfi_kif_get(sp->ifname); 237 if (kif == NULL) { 238 if (pf_status.debug >= PF_DEBUG_MISC) 239 kprintf("pfsync_insert_net_state: " 240 "unknown interface: %s\n", sp->ifname); 241 /* skip this state */ 242 return (0); 243 } 244 245 /* 246 * If the ruleset checksums match, it's safe to associate the state 247 * with the rule of that number. 248 */ 249 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && chksum_flag && 250 ntohl(sp->rule) < 251 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 252 r = pf_main_ruleset.rules[ 253 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 254 else 255 r = &pf_default_rule; 256 257 if (!r->max_states || r->states < r->max_states) 258 st = pool_get(&pf_state_pl, PR_NOWAIT); 259 if (st == NULL) { 260 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 261 return (ENOMEM); 262 } 263 bzero(st, sizeof(*st)); 264 265 if ((sk = pf_alloc_state_key(st)) == NULL) { 266 pool_put(&pf_state_pl, st); 267 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 268 return (ENOMEM); 269 } 270 271 /* allocate memory for scrub info */ 272 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 273 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) { 274 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 275 if (st->src.scrub) 276 pool_put(&pf_state_scrub_pl, st->src.scrub); 277 pool_put(&pf_state_pl, st); 278 pool_put(&pf_state_key_pl, sk); 279 return (ENOMEM); 280 } 281 282 st->rule.ptr = r; 283 /* XXX get pointers to nat_rule and anchor */ 284 285 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 286 r->states++; 287 288 /* fill in the rest of the state entry */ 289 pf_state_host_ntoh(&sp->lan, &sk->lan); 290 pf_state_host_ntoh(&sp->gwy, &sk->gwy); 291 pf_state_host_ntoh(&sp->ext, &sk->ext); 292 293 pf_state_peer_ntoh(&sp->src, &st->src); 294 pf_state_peer_ntoh(&sp->dst, &st->dst); 295 296 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 297 st->creation = time_second - ntohl(sp->creation); 298 st->expire = ntohl(sp->expire) + time_second; 299 300 sk->af = sp->af; 301 sk->proto = sp->proto; 302 sk->direction = sp->direction; 303 st->log = sp->log; 304 st->timeout = sp->timeout; 305 st->allow_opts = sp->allow_opts; 306 307 bcopy(sp->id, &st->id, sizeof(st->id)); 308 st->creatorid = sp->creatorid; 309 st->sync_flags = PFSTATE_FROMSYNC; 310 311 if (pf_insert_state(kif, st)) { 312 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 313 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 314 r->states--; 315 if (st->dst.scrub) 316 pool_put(&pf_state_scrub_pl, st->dst.scrub); 317 if (st->src.scrub) 318 pool_put(&pf_state_scrub_pl, st->src.scrub); 319 pool_put(&pf_state_pl, st); 320 return (EINVAL); 321 } 322 323 return (0); 324 } 325 326 void 327 pfsync_input(struct mbuf *m, ...) 328 { 329 struct ip *ip = mtod(m, struct ip *); 330 struct pfsync_header *ph; 331 struct pfsync_softc *sc = pfsyncif; 332 struct pf_state *st; 333 struct pf_state_key *sk; 334 struct pf_state_cmp id_key; 335 struct pfsync_state *sp; 336 struct pfsync_state_upd *up; 337 struct pfsync_state_del *dp; 338 struct pfsync_state_clr *cp; 339 struct pfsync_state_upd_req *rup; 340 struct pfsync_state_bus *bus; 341 #ifdef IPSEC 342 struct pfsync_tdb *pt; 343 #endif 344 struct in_addr src; 345 struct mbuf *mp; 346 int iplen, action, error, i, count, offp, sfail, stale = 0; 347 u_int8_t chksum_flag = 0; 348 349 /* This function is not yet called from anywhere */ 350 /* Still we assume for safety that pf_token must be held */ 351 ASSERT_LWKT_TOKEN_HELD(&pf_token); 352 353 pfsyncstats.pfsyncs_ipackets++; 354 355 /* verify that we have a sync interface configured */ 356 if (!sc || !sc->sc_sync_ifp || !pf_status.running) 357 goto done; 358 359 /* verify that the packet came in on the right interface */ 360 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) { 361 pfsyncstats.pfsyncs_badif++; 362 goto done; 363 } 364 365 /* verify that the IP TTL is 255. */ 366 if (ip->ip_ttl != PFSYNC_DFLTTL) { 367 pfsyncstats.pfsyncs_badttl++; 368 goto done; 369 } 370 371 iplen = ip->ip_hl << 2; 372 373 if (m->m_pkthdr.len < iplen + sizeof(*ph)) { 374 pfsyncstats.pfsyncs_hdrops++; 375 goto done; 376 } 377 378 if (iplen + sizeof(*ph) > m->m_len) { 379 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) { 380 pfsyncstats.pfsyncs_hdrops++; 381 goto done; 382 } 383 ip = mtod(m, struct ip *); 384 } 385 ph = (struct pfsync_header *)((char *)ip + iplen); 386 387 /* verify the version */ 388 if (ph->version != PFSYNC_VERSION) { 389 pfsyncstats.pfsyncs_badver++; 390 goto done; 391 } 392 393 action = ph->action; 394 count = ph->count; 395 396 /* make sure it's a valid action code */ 397 if (action >= PFSYNC_ACT_MAX) { 398 pfsyncstats.pfsyncs_badact++; 399 goto done; 400 } 401 402 /* Cheaper to grab this now than having to mess with mbufs later */ 403 src = ip->ip_src; 404 405 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 406 chksum_flag++; 407 408 switch (action) { 409 case PFSYNC_ACT_CLR: { 410 struct pf_state *nexts; 411 struct pf_state_key *nextsk; 412 struct pfi_kif *kif; 413 u_int32_t creatorid; 414 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 415 sizeof(*cp), &offp)) == NULL) { 416 pfsyncstats.pfsyncs_badlen++; 417 return; 418 } 419 cp = (struct pfsync_state_clr *)(mp->m_data + offp); 420 creatorid = cp->creatorid; 421 422 crit_enter(); 423 if (cp->ifname[0] == '\0') { 424 for (st = RB_MIN(pf_state_tree_id, &tree_id); 425 st; st = nexts) { 426 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st); 427 if (st->creatorid == creatorid) { 428 st->sync_flags |= PFSTATE_FROMSYNC; 429 pf_unlink_state(st); 430 } 431 } 432 } else { 433 if ((kif = pfi_kif_get(cp->ifname)) == NULL) { 434 crit_exit(); 435 return; 436 } 437 for (sk = RB_MIN(pf_state_tree_lan_ext, 438 &pf_statetbl_lan_ext); sk; sk = nextsk) { 439 nextsk = RB_NEXT(pf_state_tree_lan_ext, 440 &pf_statetbl_lan_ext, sk); 441 TAILQ_FOREACH(st, &sk->states, next) { 442 if (st->creatorid == creatorid) { 443 st->sync_flags |= 444 PFSTATE_FROMSYNC; 445 pf_unlink_state(st); 446 } 447 } 448 } 449 } 450 crit_exit(); 451 452 break; 453 } 454 case PFSYNC_ACT_INS: 455 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 456 count * sizeof(*sp), &offp)) == NULL) { 457 pfsyncstats.pfsyncs_badlen++; 458 return; 459 } 460 461 crit_enter(); 462 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 463 i < count; i++, sp++) { 464 /* check for invalid values */ 465 if (sp->timeout >= PFTM_MAX || 466 sp->src.state > PF_TCPS_PROXY_DST || 467 sp->dst.state > PF_TCPS_PROXY_DST || 468 sp->direction > PF_OUT || 469 (sp->af != AF_INET && sp->af != AF_INET6)) { 470 if (pf_status.debug >= PF_DEBUG_MISC) 471 kprintf("pfsync_insert: PFSYNC_ACT_INS: " 472 "invalid value\n"); 473 pfsyncstats.pfsyncs_badstate++; 474 continue; 475 } 476 477 if ((error = pfsync_insert_net_state(sp, 478 chksum_flag))) { 479 if (error == ENOMEM) { 480 crit_exit(); 481 goto done; 482 } 483 continue; 484 } 485 } 486 crit_exit(); 487 break; 488 case PFSYNC_ACT_UPD: 489 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 490 count * sizeof(*sp), &offp)) == NULL) { 491 pfsyncstats.pfsyncs_badlen++; 492 return; 493 } 494 495 crit_enter(); 496 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 497 i < count; i++, sp++) { 498 int flags = PFSYNC_FLAG_STALE; 499 500 /* check for invalid values */ 501 if (sp->timeout >= PFTM_MAX || 502 sp->src.state > PF_TCPS_PROXY_DST || 503 sp->dst.state > PF_TCPS_PROXY_DST) { 504 if (pf_status.debug >= PF_DEBUG_MISC) 505 kprintf("pfsync_insert: PFSYNC_ACT_UPD: " 506 "invalid value\n"); 507 pfsyncstats.pfsyncs_badstate++; 508 continue; 509 } 510 511 bcopy(sp->id, &id_key.id, sizeof(id_key.id)); 512 id_key.creatorid = sp->creatorid; 513 514 st = pf_find_state_byid(&id_key); 515 if (st == NULL) { 516 /* insert the update */ 517 if (pfsync_insert_net_state(sp, chksum_flag)) 518 pfsyncstats.pfsyncs_badstate++; 519 continue; 520 } 521 sk = st->state_key; 522 sfail = 0; 523 if (sk->proto == IPPROTO_TCP) { 524 /* 525 * The state should never go backwards except 526 * for syn-proxy states. Neither should the 527 * sequence window slide backwards. 528 */ 529 if (st->src.state > sp->src.state && 530 (st->src.state < PF_TCPS_PROXY_SRC || 531 sp->src.state >= PF_TCPS_PROXY_SRC)) 532 sfail = 1; 533 else if (SEQ_GT(st->src.seqlo, 534 ntohl(sp->src.seqlo))) 535 sfail = 3; 536 else if (st->dst.state > sp->dst.state) { 537 /* There might still be useful 538 * information about the src state here, 539 * so import that part of the update, 540 * then "fail" so we send the updated 541 * state back to the peer who is missing 542 * our what we know. */ 543 pf_state_peer_ntoh(&sp->src, &st->src); 544 /* XXX do anything with timeouts? */ 545 sfail = 7; 546 flags = 0; 547 } else if (st->dst.state >= TCPS_SYN_SENT && 548 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo))) 549 sfail = 4; 550 } else { 551 /* 552 * Non-TCP protocol state machine always go 553 * forwards 554 */ 555 if (st->src.state > sp->src.state) 556 sfail = 5; 557 else if (st->dst.state > sp->dst.state) 558 sfail = 6; 559 } 560 if (sfail) { 561 if (pf_status.debug >= PF_DEBUG_MISC) 562 kprintf("pfsync: %s stale update " 563 "(%d) id: %016jx " 564 "creatorid: %08x\n", 565 (sfail < 7 ? "ignoring" 566 : "partial"), sfail, 567 (uintmax_t)be64toh(st->id), 568 ntohl(st->creatorid)); 569 pfsyncstats.pfsyncs_badstate++; 570 571 if (!(sp->sync_flags & PFSTATE_STALE)) { 572 /* we have a better state, send it */ 573 if (sc->sc_mbuf != NULL && !stale) 574 pfsync_sendout(sc); 575 stale++; 576 if (!st->sync_flags) 577 pfsync_pack_state( 578 PFSYNC_ACT_UPD, st, flags); 579 } 580 continue; 581 } 582 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 583 pf_state_peer_ntoh(&sp->src, &st->src); 584 pf_state_peer_ntoh(&sp->dst, &st->dst); 585 st->expire = ntohl(sp->expire) + time_second; 586 st->timeout = sp->timeout; 587 } 588 if (stale && sc->sc_mbuf != NULL) 589 pfsync_sendout(sc); 590 crit_exit(); 591 break; 592 /* 593 * It's not strictly necessary for us to support the "uncompressed" 594 * delete action, but it's relatively simple and maintains consistency. 595 */ 596 case PFSYNC_ACT_DEL: 597 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 598 count * sizeof(*sp), &offp)) == NULL) { 599 pfsyncstats.pfsyncs_badlen++; 600 return; 601 } 602 603 crit_enter(); 604 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 605 i < count; i++, sp++) { 606 bcopy(sp->id, &id_key.id, sizeof(id_key.id)); 607 id_key.creatorid = sp->creatorid; 608 609 st = pf_find_state_byid(&id_key); 610 if (st == NULL) { 611 pfsyncstats.pfsyncs_badstate++; 612 continue; 613 } 614 st->sync_flags |= PFSTATE_FROMSYNC; 615 pf_unlink_state(st); 616 } 617 crit_exit(); 618 break; 619 case PFSYNC_ACT_UPD_C: { 620 int update_requested = 0; 621 622 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 623 count * sizeof(*up), &offp)) == NULL) { 624 pfsyncstats.pfsyncs_badlen++; 625 return; 626 } 627 628 crit_enter(); 629 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp); 630 i < count; i++, up++) { 631 /* check for invalid values */ 632 if (up->timeout >= PFTM_MAX || 633 up->src.state > PF_TCPS_PROXY_DST || 634 up->dst.state > PF_TCPS_PROXY_DST) { 635 if (pf_status.debug >= PF_DEBUG_MISC) 636 kprintf("pfsync_insert: " 637 "PFSYNC_ACT_UPD_C: " 638 "invalid value\n"); 639 pfsyncstats.pfsyncs_badstate++; 640 continue; 641 } 642 643 bcopy(up->id, &id_key.id, sizeof(id_key.id)); 644 id_key.creatorid = up->creatorid; 645 646 st = pf_find_state_byid(&id_key); 647 if (st == NULL) { 648 /* We don't have this state. Ask for it. */ 649 error = pfsync_request_update(up, &src); 650 if (error == ENOMEM) { 651 crit_exit(); 652 goto done; 653 } 654 update_requested = 1; 655 pfsyncstats.pfsyncs_badstate++; 656 continue; 657 } 658 sk = st->state_key; 659 sfail = 0; 660 if (sk->proto == IPPROTO_TCP) { 661 /* 662 * The state should never go backwards except 663 * for syn-proxy states. Neither should the 664 * sequence window slide backwards. 665 */ 666 if (st->src.state > up->src.state && 667 (st->src.state < PF_TCPS_PROXY_SRC || 668 up->src.state >= PF_TCPS_PROXY_SRC)) 669 sfail = 1; 670 else if (st->dst.state > up->dst.state) 671 sfail = 2; 672 else if (SEQ_GT(st->src.seqlo, 673 ntohl(up->src.seqlo))) 674 sfail = 3; 675 else if (st->dst.state >= TCPS_SYN_SENT && 676 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo))) 677 sfail = 4; 678 } else { 679 /* 680 * Non-TCP protocol state machine always go 681 * forwards 682 */ 683 if (st->src.state > up->src.state) 684 sfail = 5; 685 else if (st->dst.state > up->dst.state) 686 sfail = 6; 687 } 688 if (sfail) { 689 if (pf_status.debug >= PF_DEBUG_MISC) 690 kprintf("pfsync: ignoring stale update " 691 "(%d) id: %016llx " 692 "creatorid: %08x\n", sfail, 693 be64toh(st->id), 694 ntohl(st->creatorid)); 695 pfsyncstats.pfsyncs_badstate++; 696 697 /* we have a better state, send it out */ 698 if ((!stale || update_requested) && 699 sc->sc_mbuf != NULL) { 700 pfsync_sendout(sc); 701 update_requested = 0; 702 } 703 stale++; 704 if (!st->sync_flags) 705 pfsync_pack_state(PFSYNC_ACT_UPD, st, 706 PFSYNC_FLAG_STALE); 707 continue; 708 } 709 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 710 pf_state_peer_ntoh(&up->src, &st->src); 711 pf_state_peer_ntoh(&up->dst, &st->dst); 712 st->expire = ntohl(up->expire) + time_second; 713 st->timeout = up->timeout; 714 } 715 if ((update_requested || stale) && sc->sc_mbuf) 716 pfsync_sendout(sc); 717 crit_exit(); 718 break; 719 } 720 case PFSYNC_ACT_DEL_C: 721 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 722 count * sizeof(*dp), &offp)) == NULL) { 723 pfsyncstats.pfsyncs_badlen++; 724 return; 725 } 726 727 crit_enter(); 728 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp); 729 i < count; i++, dp++) { 730 bcopy(dp->id, &id_key.id, sizeof(id_key.id)); 731 id_key.creatorid = dp->creatorid; 732 733 st = pf_find_state_byid(&id_key); 734 if (st == NULL) { 735 pfsyncstats.pfsyncs_badstate++; 736 continue; 737 } 738 st->sync_flags |= PFSTATE_FROMSYNC; 739 pf_unlink_state(st); 740 } 741 crit_exit(); 742 break; 743 case PFSYNC_ACT_INS_F: 744 case PFSYNC_ACT_DEL_F: 745 /* not implemented */ 746 break; 747 case PFSYNC_ACT_UREQ: 748 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 749 count * sizeof(*rup), &offp)) == NULL) { 750 pfsyncstats.pfsyncs_badlen++; 751 return; 752 } 753 754 crit_enter(); 755 if (sc->sc_mbuf != NULL) 756 pfsync_sendout(sc); 757 for (i = 0, 758 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp); 759 i < count; i++, rup++) { 760 bcopy(rup->id, &id_key.id, sizeof(id_key.id)); 761 id_key.creatorid = rup->creatorid; 762 763 if (id_key.id == 0 && id_key.creatorid == 0) { 764 sc->sc_ureq_received = mycpu->gd_time_seconds; 765 if (sc->sc_bulk_send_next == NULL) 766 sc->sc_bulk_send_next = 767 TAILQ_FIRST(&state_list); 768 sc->sc_bulk_terminator = sc->sc_bulk_send_next; 769 if (pf_status.debug >= PF_DEBUG_MISC) 770 kprintf("pfsync: received " 771 "bulk update request\n"); 772 pfsync_send_bus(sc, PFSYNC_BUS_START); 773 lwkt_reltoken(&pf_token); 774 callout_reset(&sc->sc_bulk_tmo, 1 * hz, 775 pfsync_bulk_update, 776 LIST_FIRST(&pfsync_list)); 777 lwkt_gettoken(&pf_token); 778 } else { 779 st = pf_find_state_byid(&id_key); 780 if (st == NULL) { 781 pfsyncstats.pfsyncs_badstate++; 782 continue; 783 } 784 if (!st->sync_flags) 785 pfsync_pack_state(PFSYNC_ACT_UPD, 786 st, 0); 787 } 788 } 789 if (sc->sc_mbuf != NULL) 790 pfsync_sendout(sc); 791 crit_exit(); 792 break; 793 case PFSYNC_ACT_BUS: 794 /* If we're not waiting for a bulk update, who cares. */ 795 if (sc->sc_ureq_sent == 0) 796 break; 797 798 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 799 sizeof(*bus), &offp)) == NULL) { 800 pfsyncstats.pfsyncs_badlen++; 801 return; 802 } 803 bus = (struct pfsync_state_bus *)(mp->m_data + offp); 804 switch (bus->status) { 805 case PFSYNC_BUS_START: 806 lwkt_reltoken(&pf_token); 807 callout_reset(&sc->sc_bulkfail_tmo, 808 pf_pool_limits[PF_LIMIT_STATES].limit / 809 (PFSYNC_BULKPACKETS * sc->sc_maxcount), 810 pfsync_bulkfail, LIST_FIRST(&pfsync_list)); 811 lwkt_gettoken(&pf_token); 812 if (pf_status.debug >= PF_DEBUG_MISC) 813 kprintf("pfsync: received bulk " 814 "update start\n"); 815 break; 816 case PFSYNC_BUS_END: 817 if (mycpu->gd_time_seconds - ntohl(bus->endtime) >= 818 sc->sc_ureq_sent) { 819 /* that's it, we're happy */ 820 sc->sc_ureq_sent = 0; 821 sc->sc_bulk_tries = 0; 822 lwkt_reltoken(&pf_token); 823 callout_stop(&sc->sc_bulkfail_tmo); 824 lwkt_gettoken(&pf_token); 825 #if NCARP > 0 826 if (!pfsync_sync_ok) { 827 lwkt_reltoken(&pf_token); 828 carp_group_demote_adj(&sc->sc_if, -1); 829 lwkt_gettoken(&pf_token); 830 } 831 #endif 832 pfsync_sync_ok = 1; 833 if (pf_status.debug >= PF_DEBUG_MISC) 834 kprintf("pfsync: received valid " 835 "bulk update end\n"); 836 } else { 837 if (pf_status.debug >= PF_DEBUG_MISC) 838 kprintf("pfsync: received invalid " 839 "bulk update end: bad timestamp\n"); 840 } 841 break; 842 } 843 break; 844 #ifdef IPSEC 845 case PFSYNC_ACT_TDB_UPD: 846 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 847 count * sizeof(*pt), &offp)) == NULL) { 848 pfsyncstats.pfsyncs_badlen++; 849 return; 850 } 851 crit_enter(); 852 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp); 853 i < count; i++, pt++) 854 pfsync_update_net_tdb(pt); 855 crit_exit(); 856 break; 857 #endif 858 } 859 860 done: 861 if (m) 862 m_freem(m); 863 } 864 865 int 866 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 867 struct rtentry *rt) 868 { 869 m_freem(m); 870 return (0); 871 } 872 873 /* ARGSUSED */ 874 int 875 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr) 876 { 877 struct pfsync_softc *sc = ifp->if_softc; 878 struct ifreq *ifr = (struct ifreq *)data; 879 struct ip_moptions *imo = &sc->sc_imo; 880 struct pfsyncreq pfsyncr; 881 struct ifnet *sifp; 882 int error; 883 884 lwkt_gettoken(&pf_token); 885 886 switch (cmd) { 887 case SIOCSIFADDR: 888 case SIOCAIFADDR: 889 case SIOCSIFDSTADDR: 890 case SIOCSIFFLAGS: 891 if (ifp->if_flags & IFF_UP) 892 ifp->if_flags |= IFF_RUNNING; 893 else 894 ifp->if_flags &= ~IFF_RUNNING; 895 break; 896 case SIOCSIFMTU: 897 if (ifr->ifr_mtu < PFSYNC_MINMTU) { 898 lwkt_reltoken(&pf_token); 899 return (EINVAL); 900 } 901 if (ifr->ifr_mtu > MCLBYTES) 902 ifr->ifr_mtu = MCLBYTES; 903 crit_enter(); 904 if (ifr->ifr_mtu < ifp->if_mtu) 905 pfsync_sendout(sc); 906 pfsync_setmtu(sc, ifr->ifr_mtu); 907 crit_exit(); 908 break; 909 case SIOCGETPFSYNC: 910 bzero(&pfsyncr, sizeof(pfsyncr)); 911 if (sc->sc_sync_ifp) 912 strlcpy(pfsyncr.pfsyncr_syncdev, 913 sc->sc_sync_ifp->if_xname, IFNAMSIZ); 914 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 915 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 916 lwkt_reltoken(&pf_token); 917 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)))) 918 return (error); 919 lwkt_gettoken(&pf_token); 920 break; 921 case SIOCSETPFSYNC: 922 if ((error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY)) != 0) { 923 lwkt_reltoken(&pf_token); 924 return (error); 925 } 926 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) { 927 lwkt_reltoken(&pf_token); 928 return (error); 929 } 930 931 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 932 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 933 else 934 sc->sc_sync_peer.s_addr = 935 pfsyncr.pfsyncr_syncpeer.s_addr; 936 937 if (pfsyncr.pfsyncr_maxupdates > 255) { 938 lwkt_reltoken(&pf_token); 939 return (EINVAL); 940 } 941 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 942 943 if (pfsyncr.pfsyncr_syncdev[0] == 0) { 944 sc->sc_sync_ifp = NULL; 945 if (sc->sc_mbuf_net != NULL) { 946 /* Don't keep stale pfsync packets around. */ 947 crit_enter(); 948 m_freem(sc->sc_mbuf_net); 949 sc->sc_mbuf_net = NULL; 950 sc->sc_statep_net.s = NULL; 951 crit_exit(); 952 } 953 if (imo->imo_num_memberships > 0) { 954 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 955 imo->imo_multicast_ifp = NULL; 956 } 957 break; 958 } 959 960 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) { 961 lwkt_reltoken(&pf_token); 962 return (EINVAL); 963 } 964 965 crit_enter(); 966 if (sifp->if_mtu < sc->sc_if.if_mtu || 967 (sc->sc_sync_ifp != NULL && 968 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) || 969 sifp->if_mtu < MCLBYTES - sizeof(struct ip)) 970 pfsync_sendout(sc); 971 sc->sc_sync_ifp = sifp; 972 973 pfsync_setmtu(sc, sc->sc_if.if_mtu); 974 975 if (imo->imo_num_memberships > 0) { 976 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 977 imo->imo_multicast_ifp = NULL; 978 } 979 980 if (sc->sc_sync_ifp && 981 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 982 struct in_addr addr; 983 984 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) { 985 sc->sc_sync_ifp = NULL; 986 lwkt_reltoken(&pf_token); 987 crit_exit(); 988 return (EADDRNOTAVAIL); 989 } 990 991 addr.s_addr = INADDR_PFSYNC_GROUP; 992 993 if ((imo->imo_membership[0] = 994 in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) { 995 sc->sc_sync_ifp = NULL; 996 lwkt_reltoken(&pf_token); 997 crit_exit(); 998 return (ENOBUFS); 999 } 1000 imo->imo_num_memberships++; 1001 imo->imo_multicast_ifp = sc->sc_sync_ifp; 1002 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 1003 imo->imo_multicast_loop = 0; 1004 } 1005 1006 if (sc->sc_sync_ifp || 1007 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) { 1008 /* Request a full state table update. */ 1009 sc->sc_ureq_sent = mycpu->gd_time_seconds; 1010 #if NCARP > 0 1011 if (pfsync_sync_ok) 1012 carp_group_demote_adj(&sc->sc_if, 1); 1013 #endif 1014 pfsync_sync_ok = 0; 1015 if (pf_status.debug >= PF_DEBUG_MISC) 1016 kprintf("pfsync: requesting bulk update\n"); 1017 lwkt_reltoken(&pf_token); 1018 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 1019 pfsync_bulkfail, LIST_FIRST(&pfsync_list)); 1020 lwkt_gettoken(&pf_token); 1021 error = pfsync_request_update(NULL, NULL); 1022 if (error == ENOMEM) { 1023 lwkt_reltoken(&pf_token); 1024 crit_exit(); 1025 return (ENOMEM); 1026 } 1027 pfsync_sendout(sc); 1028 } 1029 crit_exit(); 1030 1031 break; 1032 1033 default: 1034 lwkt_reltoken(&pf_token); 1035 return (ENOTTY); 1036 } 1037 1038 lwkt_reltoken(&pf_token); 1039 return (0); 1040 } 1041 1042 void 1043 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req) 1044 { 1045 int mtu; 1046 1047 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req) 1048 mtu = sc->sc_sync_ifp->if_mtu; 1049 else 1050 mtu = mtu_req; 1051 1052 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) / 1053 sizeof(struct pfsync_state); 1054 if (sc->sc_maxcount > 254) 1055 sc->sc_maxcount = 254; 1056 sc->sc_if.if_mtu = sizeof(struct pfsync_header) + 1057 sc->sc_maxcount * sizeof(struct pfsync_state); 1058 } 1059 1060 struct mbuf * 1061 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp) 1062 { 1063 struct pfsync_header *h; 1064 struct mbuf *m; 1065 int len; 1066 1067 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1068 1069 MGETHDR(m, M_WAITOK, MT_DATA); 1070 if (m == NULL) { 1071 sc->sc_if.if_oerrors++; 1072 return (NULL); 1073 } 1074 1075 switch (action) { 1076 case PFSYNC_ACT_CLR: 1077 len = sizeof(struct pfsync_header) + 1078 sizeof(struct pfsync_state_clr); 1079 break; 1080 case PFSYNC_ACT_UPD_C: 1081 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) + 1082 sizeof(struct pfsync_header); 1083 break; 1084 case PFSYNC_ACT_DEL_C: 1085 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) + 1086 sizeof(struct pfsync_header); 1087 break; 1088 case PFSYNC_ACT_UREQ: 1089 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) + 1090 sizeof(struct pfsync_header); 1091 break; 1092 case PFSYNC_ACT_BUS: 1093 len = sizeof(struct pfsync_header) + 1094 sizeof(struct pfsync_state_bus); 1095 break; 1096 case PFSYNC_ACT_TDB_UPD: 1097 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) + 1098 sizeof(struct pfsync_header); 1099 break; 1100 default: 1101 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) + 1102 sizeof(struct pfsync_header); 1103 break; 1104 } 1105 1106 if (len > MHLEN) { 1107 MCLGET(m, M_WAITOK); 1108 if ((m->m_flags & M_EXT) == 0) { 1109 m_free(m); 1110 sc->sc_if.if_oerrors++; 1111 return (NULL); 1112 } 1113 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1); 1114 } else 1115 MH_ALIGN(m, len); 1116 1117 m->m_pkthdr.rcvif = NULL; 1118 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header); 1119 h = mtod(m, struct pfsync_header *); 1120 h->version = PFSYNC_VERSION; 1121 h->af = 0; 1122 h->count = 0; 1123 h->action = action; 1124 1125 *sp = (void *)((char *)h + PFSYNC_HDRLEN); 1126 lwkt_reltoken(&pf_token); 1127 callout_reset(&sc->sc_tmo, hz, pfsync_timeout, 1128 LIST_FIRST(&pfsync_list)); 1129 lwkt_gettoken(&pf_token); 1130 return (m); 1131 } 1132 1133 int 1134 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags) 1135 { 1136 struct ifnet *ifp = NULL; 1137 struct pfsync_softc *sc = pfsyncif; 1138 struct pfsync_header *h, *h_net; 1139 struct pfsync_state *sp = NULL; 1140 struct pfsync_state_upd *up = NULL; 1141 struct pfsync_state_del *dp = NULL; 1142 struct pf_state_key *sk = st->state_key; 1143 struct pf_rule *r; 1144 u_long secs; 1145 int ret = 0; 1146 u_int8_t i = 255, newaction = 0; 1147 1148 if (sc == NULL) 1149 return (0); 1150 ifp = &sc->sc_if; 1151 1152 /* 1153 * If a packet falls in the forest and there's nobody around to 1154 * hear, does it make a sound? 1155 */ 1156 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1157 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1158 /* Don't leave any stale pfsync packets hanging around. */ 1159 if (sc->sc_mbuf != NULL) { 1160 m_freem(sc->sc_mbuf); 1161 sc->sc_mbuf = NULL; 1162 sc->sc_statep.s = NULL; 1163 } 1164 return (0); 1165 } 1166 1167 if (action >= PFSYNC_ACT_MAX) 1168 return (EINVAL); 1169 1170 crit_enter(); 1171 if (sc->sc_mbuf == NULL) { 1172 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1173 (void *)&sc->sc_statep.s)) == NULL) { 1174 crit_exit(); 1175 return (ENOMEM); 1176 } 1177 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1178 } else { 1179 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1180 if (h->action != action) { 1181 pfsync_sendout(sc); 1182 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1183 (void *)&sc->sc_statep.s)) == NULL) { 1184 crit_exit(); 1185 return (ENOMEM); 1186 } 1187 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1188 } else { 1189 /* 1190 * If it's an update, look in the packet to see if 1191 * we already have an update for the state. 1192 */ 1193 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) { 1194 struct pfsync_state *usp = 1195 (void *)((char *)h + PFSYNC_HDRLEN); 1196 1197 for (i = 0; i < h->count; i++) { 1198 if (!memcmp(usp->id, &st->id, 1199 PFSYNC_ID_LEN) && 1200 usp->creatorid == st->creatorid) { 1201 sp = usp; 1202 sp->updates++; 1203 break; 1204 } 1205 usp++; 1206 } 1207 } 1208 } 1209 } 1210 1211 secs = time_second; 1212 1213 st->pfsync_time = mycpu->gd_time_seconds;; 1214 1215 if (sp == NULL) { 1216 /* not a "duplicate" update */ 1217 i = 255; 1218 sp = sc->sc_statep.s++; 1219 sc->sc_mbuf->m_pkthdr.len = 1220 sc->sc_mbuf->m_len += sizeof(struct pfsync_state); 1221 h->count++; 1222 bzero(sp, sizeof(*sp)); 1223 1224 bcopy(&st->id, sp->id, sizeof(sp->id)); 1225 sp->creatorid = st->creatorid; 1226 1227 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 1228 pf_state_host_hton(&sk->lan, &sp->lan); 1229 pf_state_host_hton(&sk->gwy, &sp->gwy); 1230 pf_state_host_hton(&sk->ext, &sp->ext); 1231 1232 bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr)); 1233 1234 sp->creation = htonl(secs - st->creation); 1235 pf_state_counter_hton(st->packets[0], sp->packets[0]); 1236 pf_state_counter_hton(st->packets[1], sp->packets[1]); 1237 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 1238 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 1239 if ((r = st->rule.ptr) == NULL) 1240 sp->rule = htonl(-1); 1241 else 1242 sp->rule = htonl(r->nr); 1243 if ((r = st->anchor.ptr) == NULL) 1244 sp->anchor = htonl(-1); 1245 else 1246 sp->anchor = htonl(r->nr); 1247 sp->af = sk->af; 1248 sp->proto = sk->proto; 1249 sp->direction = sk->direction; 1250 sp->log = st->log; 1251 sp->allow_opts = st->allow_opts; 1252 sp->timeout = st->timeout; 1253 1254 if (flags & PFSYNC_FLAG_STALE) 1255 sp->sync_flags |= PFSTATE_STALE; 1256 } 1257 1258 pf_state_peer_hton(&st->src, &sp->src); 1259 pf_state_peer_hton(&st->dst, &sp->dst); 1260 1261 if (st->expire <= secs) 1262 sp->expire = htonl(0); 1263 else 1264 sp->expire = htonl(st->expire - secs); 1265 1266 /* do we need to build "compressed" actions for network transfer? */ 1267 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) { 1268 switch (action) { 1269 case PFSYNC_ACT_UPD: 1270 newaction = PFSYNC_ACT_UPD_C; 1271 break; 1272 case PFSYNC_ACT_DEL: 1273 newaction = PFSYNC_ACT_DEL_C; 1274 break; 1275 default: 1276 /* by default we just send the uncompressed states */ 1277 break; 1278 } 1279 } 1280 1281 if (newaction) { 1282 if (sc->sc_mbuf_net == NULL) { 1283 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction, 1284 (void *)&sc->sc_statep_net.s)) == NULL) { 1285 crit_exit(); 1286 return (ENOMEM); 1287 } 1288 } 1289 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *); 1290 1291 switch (newaction) { 1292 case PFSYNC_ACT_UPD_C: 1293 if (i != 255) { 1294 up = (void *)((char *)h_net + 1295 PFSYNC_HDRLEN + (i * sizeof(*up))); 1296 up->updates++; 1297 } else { 1298 h_net->count++; 1299 sc->sc_mbuf_net->m_pkthdr.len = 1300 sc->sc_mbuf_net->m_len += sizeof(*up); 1301 up = sc->sc_statep_net.u++; 1302 1303 bzero(up, sizeof(*up)); 1304 bcopy(&st->id, up->id, sizeof(up->id)); 1305 up->creatorid = st->creatorid; 1306 } 1307 up->timeout = st->timeout; 1308 up->expire = sp->expire; 1309 up->src = sp->src; 1310 up->dst = sp->dst; 1311 break; 1312 case PFSYNC_ACT_DEL_C: 1313 sc->sc_mbuf_net->m_pkthdr.len = 1314 sc->sc_mbuf_net->m_len += sizeof(*dp); 1315 dp = sc->sc_statep_net.d++; 1316 h_net->count++; 1317 1318 bzero(dp, sizeof(*dp)); 1319 bcopy(&st->id, dp->id, sizeof(dp->id)); 1320 dp->creatorid = st->creatorid; 1321 break; 1322 } 1323 } 1324 1325 if (h->count == sc->sc_maxcount || 1326 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates))) 1327 ret = pfsync_sendout(sc); 1328 1329 crit_exit(); 1330 return (ret); 1331 } 1332 1333 int 1334 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src) 1335 { 1336 struct ifnet *ifp = NULL; 1337 struct pfsync_header *h; 1338 struct pfsync_softc *sc = pfsyncif; 1339 struct pfsync_state_upd_req *rup; 1340 int ret = 0; 1341 1342 if (sc == NULL) 1343 return (0); 1344 1345 ifp = &sc->sc_if; 1346 if (sc->sc_mbuf == NULL) { 1347 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1348 (void *)&sc->sc_statep.s)) == NULL) 1349 return (ENOMEM); 1350 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1351 } else { 1352 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1353 if (h->action != PFSYNC_ACT_UREQ) { 1354 pfsync_sendout(sc); 1355 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1356 (void *)&sc->sc_statep.s)) == NULL) 1357 return (ENOMEM); 1358 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1359 } 1360 } 1361 1362 if (src != NULL) 1363 sc->sc_sendaddr = *src; 1364 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup); 1365 h->count++; 1366 rup = sc->sc_statep.r++; 1367 bzero(rup, sizeof(*rup)); 1368 if (up != NULL) { 1369 bcopy(up->id, rup->id, sizeof(rup->id)); 1370 rup->creatorid = up->creatorid; 1371 } 1372 1373 if (h->count == sc->sc_maxcount) 1374 ret = pfsync_sendout(sc); 1375 1376 return (ret); 1377 } 1378 1379 int 1380 pfsync_clear_states(u_int32_t creatorid, char *ifname) 1381 { 1382 struct ifnet *ifp = NULL; 1383 struct pfsync_softc *sc = pfsyncif; 1384 struct pfsync_state_clr *cp; 1385 int ret; 1386 1387 if (sc == NULL) 1388 return (0); 1389 1390 ifp = &sc->sc_if; 1391 crit_enter(); 1392 if (sc->sc_mbuf != NULL) 1393 pfsync_sendout(sc); 1394 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR, 1395 (void *)&sc->sc_statep.c)) == NULL) { 1396 crit_exit(); 1397 return (ENOMEM); 1398 } 1399 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp); 1400 cp = sc->sc_statep.c; 1401 cp->creatorid = creatorid; 1402 if (ifname != NULL) 1403 strlcpy(cp->ifname, ifname, IFNAMSIZ); 1404 1405 ret = (pfsync_sendout(sc)); 1406 crit_exit(); 1407 return (ret); 1408 } 1409 1410 void 1411 pfsync_timeout(void *v) 1412 { 1413 struct pfsync_softc *sc = v; 1414 1415 crit_enter(); 1416 pfsync_sendout(sc); 1417 crit_exit(); 1418 } 1419 1420 void 1421 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status) 1422 { 1423 struct pfsync_state_bus *bus; 1424 1425 if (sc->sc_mbuf != NULL) 1426 pfsync_sendout(sc); 1427 1428 if (pfsync_sync_ok && 1429 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS, 1430 (void *)&sc->sc_statep.b)) != NULL) { 1431 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus); 1432 bus = sc->sc_statep.b; 1433 bus->creatorid = pf_status.hostid; 1434 bus->status = status; 1435 bus->endtime = htonl(mycpu->gd_time_seconds - sc->sc_ureq_received); 1436 pfsync_sendout(sc); 1437 } 1438 } 1439 1440 void 1441 pfsync_bulk_update(void *v) 1442 { 1443 struct pfsync_softc *sc = v; 1444 int i = 0; 1445 struct pf_state *state; 1446 1447 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1448 1449 crit_enter(); 1450 if (sc->sc_mbuf != NULL) 1451 pfsync_sendout(sc); 1452 1453 /* 1454 * Grab at most PFSYNC_BULKPACKETS worth of states which have not 1455 * been sent since the latest request was made. 1456 */ 1457 state = sc->sc_bulk_send_next; 1458 if (state) 1459 do { 1460 /* send state update if syncable and not already sent */ 1461 if (!state->sync_flags 1462 && state->timeout < PFTM_MAX 1463 && state->pfsync_time <= sc->sc_ureq_received) { 1464 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0); 1465 i++; 1466 } 1467 1468 /* figure next state to send */ 1469 state = TAILQ_NEXT(state, entry_list); 1470 1471 /* wrap to start of list if we hit the end */ 1472 if (!state) 1473 state = TAILQ_FIRST(&state_list); 1474 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS && 1475 state != sc->sc_bulk_terminator); 1476 1477 if (!state || state == sc->sc_bulk_terminator) { 1478 /* we're done */ 1479 pfsync_send_bus(sc, PFSYNC_BUS_END); 1480 sc->sc_ureq_received = 0; 1481 sc->sc_bulk_send_next = NULL; 1482 sc->sc_bulk_terminator = NULL; 1483 lwkt_reltoken(&pf_token); 1484 callout_stop(&sc->sc_bulk_tmo); 1485 lwkt_gettoken(&pf_token); 1486 if (pf_status.debug >= PF_DEBUG_MISC) 1487 kprintf("pfsync: bulk update complete\n"); 1488 } else { 1489 /* look again for more in a bit */ 1490 lwkt_reltoken(&pf_token); 1491 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_timeout, 1492 LIST_FIRST(&pfsync_list)); 1493 lwkt_gettoken(&pf_token); 1494 sc->sc_bulk_send_next = state; 1495 } 1496 if (sc->sc_mbuf != NULL) 1497 pfsync_sendout(sc); 1498 crit_exit(); 1499 } 1500 1501 void 1502 pfsync_bulkfail(void *v) 1503 { 1504 struct pfsync_softc *sc = v; 1505 int error; 1506 1507 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1508 1509 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 1510 /* Try again in a bit */ 1511 lwkt_reltoken(&pf_token); 1512 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulkfail, 1513 LIST_FIRST(&pfsync_list)); 1514 lwkt_gettoken(&pf_token); 1515 crit_enter(); 1516 error = pfsync_request_update(NULL, NULL); 1517 if (error == ENOMEM) { 1518 if (pf_status.debug >= PF_DEBUG_MISC) 1519 kprintf("pfsync: cannot allocate mbufs for " 1520 "bulk update\n"); 1521 } else 1522 pfsync_sendout(sc); 1523 crit_exit(); 1524 } else { 1525 /* Pretend like the transfer was ok */ 1526 sc->sc_ureq_sent = 0; 1527 sc->sc_bulk_tries = 0; 1528 #if NCARP > 0 1529 if (!pfsync_sync_ok) 1530 carp_group_demote_adj(&sc->sc_if, -1); 1531 #endif 1532 pfsync_sync_ok = 1; 1533 if (pf_status.debug >= PF_DEBUG_MISC) 1534 kprintf("pfsync: failed to receive " 1535 "bulk update status\n"); 1536 lwkt_reltoken(&pf_token); 1537 callout_stop(&sc->sc_bulkfail_tmo); 1538 lwkt_gettoken(&pf_token); 1539 } 1540 } 1541 1542 /* This must be called in splnet() */ 1543 int 1544 pfsync_sendout(struct pfsync_softc *sc) 1545 { 1546 #if NBPFILTER > 0 1547 struct ifnet *ifp = &sc->sc_if; 1548 #endif 1549 struct mbuf *m; 1550 1551 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1552 1553 lwkt_reltoken(&pf_token); 1554 callout_stop(&sc->sc_tmo); 1555 lwkt_gettoken(&pf_token); 1556 1557 if (sc->sc_mbuf == NULL) 1558 return (0); 1559 m = sc->sc_mbuf; 1560 sc->sc_mbuf = NULL; 1561 sc->sc_statep.s = NULL; 1562 1563 #if NBPFILTER > 0 1564 if (ifp->if_bpf) { 1565 lwkt_reltoken(&pf_token); 1566 bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT); 1567 lwkt_gettoken(&pf_token); 1568 } 1569 #endif 1570 1571 if (sc->sc_mbuf_net) { 1572 m_freem(m); 1573 m = sc->sc_mbuf_net; 1574 sc->sc_mbuf_net = NULL; 1575 sc->sc_statep_net.s = NULL; 1576 } 1577 1578 return pfsync_sendout_mbuf(sc, m); 1579 } 1580 1581 int 1582 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m) 1583 { 1584 struct sockaddr sa; 1585 struct ip *ip; 1586 1587 if (sc->sc_sync_ifp || 1588 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) { 1589 M_PREPEND(m, sizeof(struct ip), M_WAITOK); 1590 if (m == NULL) { 1591 pfsyncstats.pfsyncs_onomem++; 1592 return (0); 1593 } 1594 ip = mtod(m, struct ip *); 1595 ip->ip_v = IPVERSION; 1596 ip->ip_hl = sizeof(*ip) >> 2; 1597 ip->ip_tos = IPTOS_LOWDELAY; 1598 ip->ip_len = htons(m->m_pkthdr.len); 1599 ip->ip_id = htons(ip_randomid()); 1600 ip->ip_off = htons(IP_DF); 1601 ip->ip_ttl = PFSYNC_DFLTTL; 1602 ip->ip_p = IPPROTO_PFSYNC; 1603 ip->ip_sum = 0; 1604 1605 bzero(&sa, sizeof(sa)); 1606 ip->ip_src.s_addr = INADDR_ANY; 1607 1608 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP) 1609 m->m_flags |= M_MCAST; 1610 ip->ip_dst = sc->sc_sendaddr; 1611 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr; 1612 1613 pfsyncstats.pfsyncs_opackets++; 1614 1615 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) 1616 pfsyncstats.pfsyncs_oerrors++; 1617 } else 1618 m_freem(m); 1619 1620 return (0); 1621 } 1622 1623 static int 1624 pfsync_modevent(module_t mod, int type, void *data) 1625 { 1626 int error = 0; 1627 1628 lwkt_gettoken(&pf_token); 1629 1630 switch (type) { 1631 case MOD_LOAD: 1632 LIST_INIT(&pfsync_list); 1633 lwkt_reltoken(&pf_token); 1634 if_clone_attach(&pfsync_cloner); 1635 lwkt_gettoken(&pf_token); 1636 break; 1637 1638 case MOD_UNLOAD: 1639 lwkt_reltoken(&pf_token); 1640 if_clone_detach(&pfsync_cloner); 1641 lwkt_gettoken(&pf_token); 1642 while (!LIST_EMPTY(&pfsync_list)) 1643 pfsync_clone_destroy( 1644 &LIST_FIRST(&pfsync_list)->sc_if); 1645 break; 1646 1647 default: 1648 error = EINVAL; 1649 break; 1650 } 1651 1652 lwkt_reltoken(&pf_token); 1653 return error; 1654 } 1655 1656 static moduledata_t pfsync_mod = { 1657 "pfsync", 1658 pfsync_modevent, 1659 0 1660 }; 1661 1662 #define PFSYNC_MODVER 1 1663 1664 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 1665 MODULE_VERSION(pfsync, PFSYNC_MODVER); 1666 1667 1668 1669